The docking success rate was benchmarked as a percentage of correctly docked ligands for which the top-scored pose was within 2 Å RMSD from the reference ligand coordinates, for a set of protein-ligand complexes extracted from PDB. Although alternative definitions of docking success rate are available¹ we used the most widely accepted one in order to make Lead Finder benchmarks comparable with competitive software benchmarks obtained elsewhere. Additionally, to increase reproducibility all docking calculations were performed independently 20 times, and docking was considered successful only when the probability of generating the top-ranked pose with 2 Å or better accuracy was at least 0.5.

A set of 407 protein-ligand complexes was used for current docking success rate measurements. This set of complexes was combined from test sets used in original benchmarking studies of such docking software as: FlexX², Glide SP³, Glide XP⁴, Gold^5,6,7, LigandFit⁸, MolDock⁹, Surflex¹⁰. For this reason, current benchmarking study of Lead Finder docking success rate appears to be the most extensive study of such kind; moreover, the results obtained with Lead Finder can be correctly compared to competitive docking programs of interest since they were obtained on the same sets of protein-ligand complexes. All benchmarking calculations were performed in two regimes (docking and screening, see Docking Algorithm section).

The docking success rate obtained with Lead Finder on different test sets ranged from 80.0% (for GlideXP and FlexX test sets) to 96.0% (for Surflex and MolDock test sets). The data in the table below show that Lead Finder outperformed all reviewed docking software programs for which reliable original benchmarks were available. More specifically, Lead Finder successfully docked 73 structures which could not be docked by FlexX, 52 structures - for Glide SP, 52 structures - for Glide XP, 30 structures - for Gold on the original Gold test set, 13 structures - for Gold on Astex diversity test set, 9 structures - for MolDock, 21 structures - for Surflex. Example of carboxypeptidase A inhibitor, which could not be correctly docked by such programs as Glide³ or FlexX², but was successfully docked by Lead Finder, is illustrated in the figure.

	FlexX2	Glide SP3	Glide XP4	Gold5	Gold6	Gold7	LigandFit8	MolDock9	Surflex10	All
Original data	46.5	70.2	69.4	72.4	76.5	— a)	— b)	87.0	70.4	n/a
Lead Finder docking regime	85.0	82.3	81.3	87.3	90.6	92.4	87.3	96.1	96.3	85.0
Lead Finder screening regime	76.5	77.3	77.2	81.3	78.8	83.7	82.3	79.2	76.5	79.0
Number of structures	200	282	268	134	85	92	84	77	81	407

The docking success rate achieved by Lead Finder in the faster, less accurate screening regime was also higher than the original benchmarks obtained with competitive software, except MolDock. Such results were achieved by adjusting settings of the docking algorithm to minimize decrease in docking success rate while increasing the speed of calculations. The applicability of screening regime for high-performance calculations is illustrated in the Computing Speed section.

---

• J. C. Cole, C. W. Murray, J.W.M. Nissink, R. D. Taylor, R. Taylor, Comparing Protein-Ligand Docking Programs Is Difficult, PROTEINS: Structure, Function, and Bioinformatics, 2005, 60, 325-332.[ Abstract ].
• M. Rarey, B. Kramer, T. Lengauer, Multiple automatic base selection: Protein-ligand docking based on incremental construction without manual intervention J Comp Aid Mol Des, 1997, 11, 369-384. [ Abstract ].
• R. A. Friesner, R. B. Murphy, M. P. Repasky, L. L. Frye, J. R. Greenwood, T. A. Halgren, P. C. Sanschagrin, D. T. Mainz, Glide: A New Approach for Rapid, Accurate Docking and Scoring. 1. Method and Assessment of Docking Accuracy, J Med Chem 2004, 47, 1739-1749.[ Abstract ].
• R. A. Friesner, J. L. Banks, R. B. Murphy, T. A. Halgren, J. J. Klicic, D. T. Mainz, M. P. Repasky, E. H. Knoll, M. Shelley, J. K. Perry, D. E. Shaw, P. Francis, P. S. Shenkin, Extra Precision Glide: Docking and Scoring Ltdorporating a Model of Hydrophobic Enclosure for Protein-Ligand Complexes, J Med Chem, 2006, 49, 6177-6196.[ Abstract ].
• G. Jones, P. Willett, R. C. Glen, A. R. Leach, R. Taylor Development and Validation of a Genetic Algorithm for Flexible Docking, J Mol Biol, 1997, 267, 727-748. [ Abstract ].
• M. J. Hartshorn, M. L. Verdonk, G. Chessari, S. C. Brewerton, W.T..M. Mooij, P. N. Mortenson, C. W. Murray, Diverse, High-Quality Test Set for the Validation of Protein-Ligand Docking Performance, J. Med Chem, 2007, 50, 726-741. [ Abstract ].
• J.W.M. Nissink, C. Murray, M. Hartshorn, M. L. Verdonk, J. C. Cole, R. Taylor A NewTest Set for Validating Predictions of Protein-Ligand Interaction, PROTEINS: Structure, Function, and Genetics, 2002, 49, 457-471.[ Abstract ].
• C. M. Venkatachalam, X. Jiang, T. Oldfield, M. Waldman LigandFit: a novel method for the shape-directed rapid docking of ligands to protein active sites, J Mol Graph Model, 2003,21, 289-307. [ Abstract ].
• R. Thomsen, M. H. Christensen, MolDock: A new technique for high-accuracy molecular docking, J Med Chem, 2006, 49, 3315-3321. [ Abstract ].
• A. N. Jain, Surflex: Fully Automatic Flexible Molecular Docking Using a Molecular Similarity-Based Search Engine, J Med Chem, 2003, 46, 499-511.[ Abstract ].